These are variables in models or theoretical distributions that typically need to be fitted to data or provided with values based on other knowledge. For example, the Binomial distribution has two parameters – Binomial(N.p). One parameter, N, is the number of trials and the other parameter, p, the probability of success of each (independent) trial. If we use this for 10 tosses of a fair coin, we can fill in these parameters based on our knowledge: N=10 and p=0.5. However, if we are uncertain whether the coin is biased, we may toss it a hundred times (N=100), but use the fact that it is a binomial distribution to fit p, that is to estimate the bias of the coin.
In a more Human–Computer Interaction-specific example, Fitts' Law says that time to hit a target is A + B x IoD (where IoD is the index of difficulty (IoD)), but the parameters A and B vary on the characteristics of the device used (e.g. mouse vs trackpad vs finger pointing).
Used on page 110